Imaging saline tracer infiltration into unsaturated sandy soil using full-waveform inversion of cross-borehole ground penetrating radar

Abstract:

Cross-borehole ground penetrating radar (GPR) can provide high-resolution (tens of centimeter) information of the subsurface between boreholes located 5-10 m apart. The method is minimal invasive and therefore provides a unique opportunity to image subsurface variability not possible with standard point-scale equipment, such as TDR- and/or capacitance probes.

Full-waveform inversion (FWI) of cross-borehole GPR uses the entire waveform of the transmitted electromagnetic signal. The recorded data contains information on the travel time of the pulse, as well as the attenuation, resulting in moisture content and electrical conductivity images of the subsurface using just one method. Few case studies of cross-borehole GPR FWI using real data have been published to date. The majority of these studies focus on estimating the variation in porosity in the saturated zone (e.g. in gravel aquifers, fractured metamorphic rock, and heterogeneous chalk sediments).

In this study, we use cross-borehole GPR to monitor the infiltration of a saline tracer into an unsaturated sandy soil. In September 2011, saline water was added across a 142 m² area at an agricultural field site in Denmark. A total of 3.3 mm saline water was applied mimicking a natural infiltration event. During the following year, the tracer infiltration into the subsurface was monitored using cross-borehole GPR at weekly to monthly intervals. Furthermore, five cores were extracted within the field site to obtain independent profiles of soil moisture and pore water conductivity for comparison.

The cross-borehole GPR data were inverted using ray-based and FWI techniques. For the FWI an appropriate starting model and an effective wavelet must be estimated. Preliminary results indicate that the data modeled for the FWI results mimic better the measured data compared to the ray-based results. However, more research is needed to investigate the influence of the used starting model and the effective wavelet estimation.